Interleaved convolution was missing bias :-( Large convolutions are 10% faster.
OpenCL implementation was missing bias :-( .
Working on test cnnalgo.dpr
Updating testcnnalgo
Bug fixing in regards to interleaved convolution.
testcnnalgo now compiles again.
Making convolutions a bit faster.
Making CAI more portable.
Preparing code for future (faster) dot product engines.
Preparing more potent dot product engines.